Achieving optimal trade-off for student dropout prediction with multi-objective reinforcement learning

Area of Science:

• Educational technology
• Machine learning in education
• Data science for learning analytics

Background:

• Student dropout prediction (SDP) is crucial for timely interventions.
• Traditional SDP methods struggle to balance prediction accuracy and earliness.
• Existing approaches often lead to sub-optimal interventions for at-risk students.

Purpose of the Study:

• To develop a novel method for optimizing the trade-off between prediction accuracy and earliness in SDP.
• To address the limitations of existing SDP techniques.
• To enhance the effectiveness of interventions for at-risk students.

Main Methods:

• Framing SDP as a partial sequence classification problem using a multiple-objective Markov decision process (MOMDP).
• Employing a vectorized reward function to maintain objective distinctiveness and enable nuanced optimization.
• Utilizing an advanced envelope Q-learning technique for comprehensive solution space exploration and Pareto-optimal strategy identification.

Main Results:

• The proposed MORL model demonstrates superior performance on real-world MOOC datasets.
• The model effectively optimizes the trade-off between prediction accuracy and earliness.
• Identified Pareto-optimal strategies cater to a broader range of user preferences.

Conclusions:

• The novel MORL approach represents a significant advancement in student dropout prediction.
• This method offers a more effective strategy for balancing accuracy and earliness in SDP.
• The findings pave the way for more informed and timely interventions in educational settings.